CCDs

Donna KubikSpring, 2005

References1. “Fully Depleted, Back-Illuminated Charge-Coupled Devices

Fabricated on High-Resistivity Silicon”, S. E. Holland, et. al., IEEETransactions on Electron Devices, Vol. 50, No.1, January (2003)225-238

2. “Determination of the electron-hole pair creation energy forsemiconductors from the spectral responsivity of photodiodes”,F. Scholze, et. al., Nuclear Instruments and Methods in PhysicsResearch A 439 (2000) 208-215

3. CCD Astronomy, Christian Buil, Willmann-Bell, 1991

4. Scientific Charge-Coupled Devices, J. R. Janesick, BellinghamWA, SPIE Press, 2001

Charge coupled device (CCD)• The CCD was developed in

1970 by Boyle and Smith atBell Labs

• They were attempting todesign a new kind ofsemiconductor memory forcomputers.

• At the same time they werelooking for a way to develop asolid-state camera for use invideo phones.

CCDs• CCDs are dynamic devices

that move charge along apredetermined paths undercontrol of clock pulses

• They now have applications inmemories, logic functions,signal processing, and imaging

CCDs• Like all photodetectors, CCDs

can be used to detect photonsor particles

– CCDs are used forastronomical imaging

– CCDs can be used asparticle detectors (i.e.vertex detectors)

OUTLINE• How does a CCD work?

• What are the basic characteristics of a CCD?

• What are the specific characteristics of the LBNL CCDs?

• What about CCDs as a vertex detector for the LCD?

CCD structure• MOS capacitor

– metal-oxide-semiconductor

• Most common to use P-typesubstrate

• Holes are majority carriers inP-type

• There are a few electrons fromthermal energy (minoritycarriers)

oxide

CCD operation• Bias gate (typical 10 V)

• Majority carriers (holes) arepushed back into interior ofsubstrate

• Zone almost free of majoritycarriers is then created at theSiO2-Si interface (depletionregion)

oxide10V

Charge injection• e-h pairs created in the

depletion region are separatedby the potential

• The e- accumulate near theSiO2-Si interface forming ann-channel

• This is also called theinversion layer

• The inversion layer carries theinformation

oxide10V

Charge transfer• After the integration time has

elapsed, the charges are readout

• Two phase, three phase, fourphase

• Example shown is the mostcommon, three phase transfer

CCD layout

CCD characteristics• Quantum efficiency• Spectral response• Transfer efficiency• Spatial resolution• Linearity• Blooming• Dark current• Sensitivity to cosmic rays• Electroluminescence• Cosmetic defects• Noise

Quantum efficiency (QE)• Quantum efficiency

• QE is generally <1.

• QE ranges from 40% to 80% for CCDs, compared to 2% to 4% forphotographic emulsion.

el/secondphtons/pixincident ofnumber average

xel/secondphotons/pi detected ofnumber average=QE

Detective Quantum Efficiency(DQE)

• If a photon produces a photoelectron, the photoelectron will notnecessarily be measured at the detector’s output.

• The photoelectron could be lost in the transfer.

• So another measure, equivalent quantum efficiency or detectivequantum efficiency (DQE) is often used.

Charge Collection Efficiency (CCE)

• QE can be > 1 for high energy photons.

• For example, a thinned CCD receiving a 1216 A photon averages3 e-h pairs.

• To take this phenomenon into account, a new quality criterion wasestablished to measure CCD performance when lit by energeticradiation (EUV and X-ray).

• This is called charge collection efficiency (CCE)

• SHOW GRAPH IN Scholze ARTICLE and next slide

Pair creation in Si• Pair creation energy is a function of the incident photon (or

particle) energy, E

• E < 1.14 eV pass through the CCD; i.e. Si is transparent in thefar-infrared

• 1.1 eV < E < 3.1 eV (1126 nm to 400 nm) will generate a singlee-h pair

• E > 3.1 eV will produce multiple e-h pairs when the energeticconduction band e- collides with other valence e-

• E> 10 eV:

temproomat Sifor eV 3.65 is which pair,

h-ean generate torequiredenergy theis

yield quantum ideal theis where

)(

he

i

he

i

E

E

eVE

!

!

=

"

"

Hysteresis• Local variations in QE may depend on past lighting (hysteresis).

• If the CCD was recently lit by intense light, those areas that werestrongly lit have higher QE.

• To minimize this effect, can illuminate the entire array with a700 nm source and then take many rapid readouts prior toexposure.

Spectral response• Spectral response close to that of Si.

• Without any special “tinkering”, the spectral response is close tothat of Si; the CCD can usually record radiation from 0.4 um to 1um.

• The possibility of recording information beyond 0.7 um is new andimportant, for the eye and photographic plates are insensitivebeyond that limit.

• The spectral sensitivity may also vary from pixel to pixel.

Transfer efficiency• During the transfer from one

stage to another, a certainnumber of charges is leftbehind

• A measure of the efficiencyis termed charge transferefficiency (CTE)

transferafter the gate following

under the charges ofnumber theis

gate aunder charges ofnumber theis

1

0

0

0

t

t

N

N

N

NNCTE

!!=

Transfer efficiency• Only a slight inefficiency is tolerable

• If 1% of the charges is left behind after each transfer, a packetinitially containing 100 electrons will contain only 37 electrons

• Today’s large arrays, a packet must be moved an average of 500times!

• CCD mfg are trying to reach a near-perfect rate

• Typical efficiency is 0.999990

electrons 37)99.0(100 100=!

"e

Transfer efficiency• CTE = f(temperature)

– It temperature is too low, carrier mobility decreases and blocksoperation of the CCD

• CTE = f(clock rate)– In astronomy, it is always possible to read the CCD slowly

• CTE = f(number of charges)– If <1000 e-, noticeable trapping– Pro: Saturate traps with LED flash (thermal current can also

do this if only moderately cooled)– Con: Adds noise

Spatial resolution• Pixels 10-30 um long

– Compare to 5 um grain size in emulsion

• Transfer inefficiency can dilute the charges of a pixel into thefollowing pixels

• Diffusion of charges in substrate during integration– Especially for long wavelengths which have longer absorption

lengths– Solution is to thin the P doped layer (substrate) so that they

recombine in the supporting substrate– But this decreases red sensitivity

Linearity• The linearity of a CCD is very good

• Linearity is limited to a working range of intensity

• For low illumination, linearity is no longer possible because ofthreshold phenomena, as charge trapping

• At high light levels, the detector becomes saturated– Space charge occurs in which e- repel each other an leak over

into adjacent pixels causing blooming (next slide)

Blooming vs. antiblooming• Anti-blooming gate designed

to bleed off overflow from asaturated pixel.

• Without this feature, a brightstar which has saturated thepixels will cause a verticalstreak.

Example of blooming

Dark Current• Even if the CCD is in complete darkness, a signal is observable

• The signal is due to thermally created charges

• Main reason to cool a CCD is to minimize the dark current

• Production rate is such that, at room temperature, a standardCCD in darkness is saturated after only a few seconds ofintegration

• Dark current becomes nearly negligible at about -100 deg C

• Thermal signal decreases by a factor of 2 for a 7 deg C drop inoperating temperature

Dark Current• But if a CCD is cooled to a very low temperature, the Si

characteristics change

– In practice, a CCD is never cooled below -120 deg C• Decreased mobility

– At low temperatures, the spectral sensitivity changes

– A shift toward blue on the infrared side of response curve• Typical 2.5 Angstroms/deg C

• The thermal current comes from the Si-SiO2 interface, sodesigners are trying to build CCDs where this effect is minimized(MPP, multi-pinned phase)

Sensitivity to cosmic rays• Cosmic rays can interact with Si

• If it arrives on an angle, wouldleave a streak

• Increased problem at highaltitudes and in space

• Can discriminate CR bycomparing images of same field

Cosmic ray

Electroluminescence• The output amplifier can emit a weak light by electroluminescence

• Solution is to turn off or lower amplifier voltage during exposureand then just turn it on immediately before readout

Cosmetic defects• A cosmetic defect is any defect locally affecting the quality of the

sensitive surface

• Some pixels may have greater dark current than their neighbors(hot spots), a completely insensitive pixel (dark spot), orsometimes a defective column.

• Most CCDs have some defects

• Mfg offer several levels of quality

• 5 or 6 bad pixels uniformly distributed in an image or even adefective column can be tolerated and corrected during imageprocessing

Noise• There are many sources of noise

– Thermal– Reset noise– Phonon (shot) noise– Readout noise

Correlated double sampling (CDS)

• Correlated double sampling yields the best representation of thetrue charge associated with each pixel

• From an electronics standpoint, there are different methods foraccomplishing this, such as digital, analog sample and hold,integration, and dual slope.

Correlated double sampling (CDS)

• Before the charge of each pixel is transferred to the output node ofthe CCD, the output node is reset to a reference value

• The pixel charge is then transferred to the output node. The finalvalue of charge assigned to this pixel is the difference between thereference value and the transferred charge

Dark noise• Although the dark count can

be subtracted, the dark noisecannot. The dark noise isapproximately the square rootof the dark count. In thisexample, the dark noise isabout 55 electrons.

Astronomical imaging• LBNL has developed red-

extended CCDs for highredshift surveys (SupernovaeCosmology Project, SNAP,DES)

• Extended red response isextremely important due to theuse of distant, high redshiftsupernovae for thedetermination of cosmologicalparameters and for deepsurveys

• Detection and follow-upspectroscopy of high redshiftobjects would greatly benefitfrom CCDs with improvednear-infrared response.

Astronomical imaging• The large focal planes of

astronomical telescopesrequire high-QE, large-formatCCD detectors.

• In order to achieve high QE,the standard scientific CCD isthinned and back illuminated

Frontside vs.backside• Back-illuminated CCDs have

exceptional quantumefficiency compared to front-illuminated CCDs

• To make a back illuminatedCCD, take a front-illuminated CCD, thin it to~20 µm, and mount it upsidedown on a rigid substrate

• The incoming light now hasa clear shot at the pixelwells without the gatestructures blocking the light

Thinning• Thinning is required, because

the relatively low-resistivitysilicon used to fabricatescientific CCDs limits the depthof the depletion region.

• If left thick, the shorter-wavelength, short absorption-length photons will beabsorbed before reaching thedepletion region where the e-hpairs they create are moreefficiently collected

Back-illumination• However, this process degrades

red and near-infrared responsesdue to the rapid increase inabsorption length in silicon atlong wavelengths (next slide)

• In addition, fringing patterns dueto multiply-reflected light areobserved in uniformlyilluminated images taken atnear-infrared wavelengthswhere the absorption lengthexceeds the CCD thickness.

• Absorption length is thereciprocal of the absorptioncoefficient (next slide)

Typical Q.E. curves for front- and back-illuminated CCDs

1.240.0020.011000

1.550.0050.1800

2.070.250.5600

3.104.05.0400

λ(nm)

Absorption coefficient• The intensity of the photon flux, F, varies as a function of the

thickness, z, of the silicon, following the relation

where α is the coefficient of intrinsic absorption of the silicon andF(0) is the incident flux on the silicon. Note:

zeFzF

!"= )0()(

seVxh !="15

10136.4

KT

um

300

)( 1

=

!"

KT

um

77

)( 1

=

!" Energy (eV)

!hcE =

Absorption length• 90% of the radiation penetrating the Si is absorbed after a distance

1

%902

!= "z

!

!

!

!

!

3.2

3.2

)ln()1.0ln(

1.0)0(

)(

,1)0(

)0()(

%90 =

"="

=

==

=

=

"

"

"

z

z

e

eF

zF

Ftognormalizin

eFzF

z

z

z

Absorption length• Notice that blue photons (400 nm) are stopped at less than 1 um,

while infrared photons (1000 nm) can cover more than 100 um.• Also note less temperature dependence at shorter wavelengths

Absorption length• Si is an indirect-bandgap

semiconductor, therefore tworegimes can be seen withchange occurring at about 2.5eV (500 nm)

• The direct bandgap energy ofSi is 2.5 eV

• The indirect bandgap energyof Si is 1.12 eV

• Above 2.5 eV, the transitionsare direct

• Below 2.5 eV, the transitionsare indirect

Direct vs. indirect bandgapmaterials

Ge is a direct bandgap material Si is an indirect bandgap material

Indirect gap

phonon

Direct vs. indirect transitions in Si

• Above 2.5 eV, the transitionsare direct

• Below 2.5 eV, the transitionsare indirect

Si is an indirect bandgap material

phonon

Direct transitions• For photon energies above

2.5 eV, absorption is highlyefficient and the absorptioncoefficient is determined byavailable conduction bandstates.

• Absorption coefficient for directtransitions, αd, varies assquare root of energy as perthe energy dependence of theconduction band density ofstates

• Temperature dependence isweak in direct bandgap regime

tempabsolute is

constant a is

)(,

T

A

TEhA directgd != "#

Direct

Indirect transitions• Below 2.5 eV, phonons are

required for momentumconservation

• Absorption less efficient

• More sensitive to temperaturedue to phonon statistics

• Absorption coefficient forindirect transitions, αi

• The 2 terms are due to phononabsorption and emission

Indirect

Astronomical imaging• The CCD developed by LBNL

achieves high QE in the redand near-infrared by using athick depleted region madepossible by the use of a high-resistivity silicon substrate

• The high resistivity allows forfully-depleted operation (200-300um) at reasonable voltages

QE of LBNL CCDTypical Q.E. curves for front- and back-illuminated CCDs

0

10

20

30

40

50

60

70

80

90

100

300 400 500 600 700 800 900 1000 1100

Wavelength (nm)

Qu

an

tum

Eff

icie

ncy (

%)

LBNL CCD QE

Much-improved IRsensitivity

Drawbacks of thick CCDs• There are several drawbacks

to a thick CCD

• Cosmic ray and terrestrialradiation sources will affectmore pixels (more volume)

• More volume for dark current

• Depth of focus issues

• Light incident at large anglesfrom the normal

LBNL remedies to drawbacks• LBNL CCD is p-channel (due

to LBNO experience withfabrication low dark currentp-i-n diodes)– Few e-/hour

• Degraded readout speed dueto the lower mobility inp-channel is not of a concernfor astronomy applications dueto relatively low readout rates

• P-channel under study forspace applications due toexpected higher resistance tocosmic ray protons

LBNL CCD vs. Commercial CCD

• Imaging at 1 um with high QE and negligible fringing is a uniquefeature of a fully depleted, thick CCD.

LBNL CCD vs. Commercial CCD• The main difference in the images is the detection of background stars in

the 1 um exposure, which are not seen in the image of Fig. 11 due toabsorption of the visible wavelengths of the background light in the dust andgas in the vicinity of the nebula.

Are CCDs a viable vertexdetector for a cold ILC?

8.3 msto

readout CCD

950 us

Compare warm and cold timing

2820bunches 200 ms

to readout CCD

337 ns

2820bunches

1.4 ns

266 ns

192bunches

192bunches

192bunches

COLD

WARM

Not enough time to readout CCD between bunches in either case

CCD readout time 50us-250us

Compare warm and cold timing• CCD readout time 50us-250us

• There is not enough time betweenbunches to readout the CCD (337ns,1.4n). Switch to CMOS pixeltechnology?

• There is enough time to readout theCCD between trains (200ms, 8ms)

• But for cold technology, that wouldmean integrating over 2820 bunchcrossings, which is far too many to sortout events!

• For warm technology, would have tointegrate over “only” 192 bunchcrossings, which is (would have been)permissible.

cycle

ms Hz @

train

ms

bunch

ns

train

bunches 2005

13372820 !=

cycle

ms Hz @

train

ns

bunch

ns

train

bunches 8120

2684.1192 !=

Cartoon of bunches/timing on next slide

COLD

WARM

Why not decrease SRF pulsewidth?

• SRF cavities take a long time to fill/drain (TTF 500us)

• Assume you’d want to be on at least as long as you are filling.Make it 1000us

• Why not longer than 1000us? Limited by rf gun and klystrons andmodulators

• NRF has high peak power, leads to shorter pulses

Why not increase SRF freq• Increase f, increase R

• Increase f, increase BCS losses

• So, R is proportional to f2

• Increase freq, must decrease temperature